Parallel Simulation
This section describes how to use the parallel simulation kernel, including:
- Parallel Simulation Concepts
- Estimating the Benefit of Parallel Simulation
- Working with the Parallel Simulation Kernels
- Running Parallel Simulations
- Debugging Parallel Simulations
- Making Models Parallel-Safe
Not all simulations can benefit from parallel execution. In fact, simulations that are not well suited for it can run more slowly in parallel, because of the additional synchronization required by parallel execution. In general, two classes of models can benefit from execution in parallel:
- Models with heavy calculations or otherwise heavy events (such as complex routing table re-calculations or radio propagation that takes terrain features into account)
- Models with a large number of objects, with enough inherent parallelism to keep the CPU pipelines filled with light events.
Another thing to keep in mind is that most of the Riverbed standard model library is not MT-safe and thus is executed sequentially even with the parallel kernel. Having any sequential events in a simulation is a major bottleneck to parallel simulation. For example, if half of events can be executed in parallel and half must be executed sequentially, the maximum theoretical speedup is 1.3x on a 2 CPU machine or 2x on a machine with infinite number of CPUs.
For help deciding if a simulation is a good candidate for parallel execution, see Estimating the Benefit of Parallel Simulation.
Note: Only network models that have been designed for parallel simulation (that is, that have been declared parallel-safe by the developer) should be run with the parallel simulation kernel.